WO2007049643A1

WO2007049643A1 - Echo suppressing method and device

Info

Publication number: WO2007049643A1
Application number: PCT/JP2006/321267
Authority: WO
Inventors: Osamu Hoshuyama
Original assignee: Nec Corporation
Priority date: 2005-10-26
Filing date: 2006-10-25
Publication date: 2007-05-03
Also published as: JP4702371B2; CN101346895A; KR100974370B1; JPWO2007049643A1; US8811627B2; EP1942582A4; CN101346895B; EP1942582A1; KR20080066049A; US20090041263A1; EP1942582B1

Abstract

A coefficient generating section (200) generates a crosstalk coefficient, which is a predetermined constant, used for calculating the echo crosstalk. A converting section (100) uses either the output signal from a sound collector or the signal which is the remainder of the subtraction of the output signal from the sound collector from the output signal of an echo canceller as a first signal, corrects the first signal according to the crosstalk coefficient generated by the coefficient generating section (200), and generates a near-end signal produced by removing the echoes from the first signal.

Description

Echo suppression method and apparatus

Technical field

TECHNICAL FIELD [0001] The present invention relates to an echo suppression method and apparatus for suppressing echo generated when loudspeaker sound and sound collection by a microphone are performed simultaneously. Background art

FIG. 1 is a block diagram showing a configuration of an echo suppressor of a first conventional example.

FIG. 1 shows an example of the configuration of an echo suppressor for suppressing echo generated in a hands-free telephone.

In FIG. 1, a voice signal (hereinafter referred to as a far end signal) of a call partner inputted from the input terminal 10 is amplified as far end voice from the speaker 2. On the other hand, for example, a voice of a speaker (hereinafter referred to as a near-end voice) is collected into the microphone 1 and an unnecessary far-end voice amplified by the speaker 2 is input. The sound input from the speaker 2 to the microphone 1 is called “eco-ichi”. The sound transmission system up to the output signal of the far-end signal force microphone 1 is called an echo path. The sound transmission system includes a speaker 2 and a microphone 1.

[0005] Only near-end speech is desired to be output from the output terminal 9 as a near-end signal, and unnecessary far-end speech included in the near-end signal is removed. In particular, if the near-end signal contains a large far-end voice signal component, the other party will hear the delayed far-end voice as an echo, making the call difficult. In order to deal with such problems, conventional techniques have adopted a method that removes echoes from both the near-end signal force and linear echo canceller. The linear echo canceller is described, for example, in Non-Patent Document 1 (Paper by Eberhard HANSLER "The hands-free telepnone problem: an annotated bibliography update", annals of telecommunications ", 1994, p360-367).

[0006] The linear echo canceller 3 estimates the transfer function of the echo path (echo path estimation), and based on this estimated transfer function, the echo signal input to the microphone 1 from the input signal (far end signal) of the speaker 2 Generate a simulated signal (echo replica signal).

[0007] The echo replica signal generated by the linear echo canceller unit 3 is input to the subtractor 4, The subtractor 4 subtracts the echo replica signal from the output signal force of the microphone 1, and the signal component of the near-end speech is extracted.

The voice detection unit 5 receives the output signal of the microphone 1, the output signal of the linear echo canceller 3, the output signal of the subtractor 4, and the far-end signal, and these signal strengths Is detected, and the detection result is output to the linear echo canceller 3.

In order to control the operation of the linear echo canceller 3, the voice detection unit 5 outputs “0” as the voice detection result when the output signal power of the microphone 1 also detects the near-end voice, and outputs an extremely small value. When a near-end voice is not detected, a large value is output.

FIG. 2 is a block diagram showing a configuration example of the linear echo canceller shown in FIG.

As shown in FIG. 2, the linear echo canceller 3 includes an adaptive filter 30 that is a linear filter and a multiplier 35. As the adaptive filter 30, various filters such as FIR type, IIR type, and lattice type are used.

The adaptive filter 30 filters the far-end signal input from the terminal 31 and outputs the processing result from the terminal 32 to the subtracter 4. The adaptive filter 30 updates the filter coefficient using a predetermined correlation operation so that the output signal of the subtractor 4 input from the terminal 33 is minimized. Therefore, the adaptive filter 30 operates so that a component having a correlation power S with the far-end signal in the output signal of the subtractor 4 is minimized. That is, the output signal force echo (far end speech) of the subtractor 4 is removed.

[0013] By the way, when the adaptive filter 30 updates the filter coefficient in a state where the output signal of the microphone 1 includes near-end speech, the ability to remove echo may be reduced due to fluctuations in the filter coefficient.

The multiplier 35 is provided for controlling the update of the filter coefficient by the adaptive filter 30, multiplies the output signal of the subtractor 4 and the output signal of the sound detection unit 5, and the result of the operation is applied to the adaptive filter. Output to 30. When the output signal of the microphone 1 includes near-end speech, the output signal of the speech detector 5 is “0” or an extremely small value as described above, so that the update of the filter coefficient by the adaptive filter 30 is suppressed. The fluctuation of the filter coefficient is reduced. As a result, a reduction in echo removal capability is suppressed. [0015] As described above, the echo suppression device of the first conventional example removes the echo of the far-end signal by using the adaptive filter.

Next, a second conventional example of an echo suppressor will be described.

[0017] The echo suppressor of the second conventional example is configured to correct a pseudo-echo (echo replica signal) used for echo suppression according to the angle of the hinge portion in the folding cellular phone device. Such a configuration is described in, for example, JP-A-8-9005.

[0018] The echo suppressor of the second conventional example detects the angle of the hinge portion, outputs a control signal corresponding to the angle, and suppresses the echo based on the control signal! And an echo control unit.

[0019] The echo control unit holds a plurality of preset echo path tracking coefficients in order to generate a pseudo echo corresponding to an echo path that varies depending on the angle of the hinge unit, and a control signal generation unit A coefficient selection circuit that selects the echo path tracking coefficient using the control signal output from the address signal and a pseudo echo correction signal for correcting the pseudo echo based on the echo path tracking coefficient selected by the coefficient selection circuit. An adaptive control circuit for outputting, a pseudo echo generating circuit for generating a pseudo echo based on the pseudo echo correction signal, and a subtracting circuit for reducing the output signal power of the voice input unit (microphone) from the generated pseudo echo are provided.

Next, an echo suppression device of a third conventional example will be described.

[0021] An echo suppression device of a third conventional example is a technique described in, for example, Japanese Patent Application Laid-Open No. 2004-056453. In the echo suppressor of the third conventional example, either the output signal of the microphone (sound collector) or the signal obtained by subtracting the output signal of the sound collector is the first signal, and the echo is suppressed. When the output signal of the canceller is the second signal, the amount of leakage of the second signal (far end signal, echo) that leaks into the first signal (near end signal) is estimated, and based on this estimation result Correct the first signal.

[0022] The estimated value of the amount of echo leakage includes an amount corresponding to the amplitude or power of the second signal during a period in which near-end speech is not detected, and an amount corresponding to the amplitude or power of the first signal. The ratio is used. In the echo suppressor of the third conventional example, an estimated value of echo leakage is calculated from the first signal and the second signal for each frequency component of the first signal and the second signal. The first signal is corrected based on the calculated estimated value.

By the way, in the echo suppression devices of the first conventional example and the second conventional example described above, echoes can be sufficiently suppressed when nonlinear elements such as distortions in the echo path are small. However, in an actual device, a speaker or the like has a large nonlinear element. The transfer function of the echo path including distortion is nonlinear, and the accurate transfer function of the echo path cannot be simulated by the linear echo canceller 3. In particular, when a loud sound is emitted from a small speaker used in a mobile phone device or the like, since the distortion of the sound is large, the echo is suppressed only by about 20 dB. In this case, the echo is transmitted as a near-end signal and can be heard by the other party's speaker, making it difficult to talk.

On the other hand, in the third conventional example, the echo is sufficiently suppressed even if the distortion of the echo path is large. However, in the echo suppressor of the third conventional example, the amount of calculation increases because the process of estimating the amount of echo leakage is complicated. In particular, the calculation amount of division increases. In addition, since the sound detection result that detects whether or not the near-end sound is included in the output signal of the microphone is used, if there is an error in the sound detection result, there is a large error in the estimated value of echo leakage. Is generated, and the corrected signal of the first signal corrected based thereon is deteriorated. That is, the echo is not sufficiently suppressed, or a large distortion occurs in the near-end speech. In particular, when used in an environment where a large amount of noise (near-end noise) is input together with the near-end speech, there is a high possibility that the error in the speech detection result will increase, so that the echo is not sufficiently suppressed or the near-end speech is large. Distortion occurs.

Disclosure of the invention

[0025] Accordingly, an object of the present invention is to provide an echo suppression method and apparatus capable of simply and sufficiently suppressing echo even when distortion caused by an echo path is large.

[0026] Another object of the present invention is to provide an echo suppression method and apparatus capable of suppressing echoes without being affected by near-end noise.

[0027] In order to achieve the above object, according to the present invention, either the output signal of the sound collector or the output signal power of the sound collector, which is obtained by subtracting the output signal of the echo canceller, is used as the first signal. When the output signal of the canceller is the second signal, the first signal is used to calculate the amount of leakage of the second signal that leaks into the first signal, using the leakage coefficient that is a preset value. Correct the signal.

[0028] When the echo canceller is a linear echo canceller, the harmonic component contained in the far-end signal appears almost as it is in the output of the echo canceller. Further, even if this echo canceller is a nonlinear echo canceller, the output of the echo canceller includes not only a few harmonic components contained in the far-end signal.

[0029] On the other hand, the output signal of the sound collector (microphone) includes the harmonic component generated by the echo of the far end signal due to the acoustic coupling between the sound collector and the loudspeaker and the distortion of the acoustic system. The ratio of these harmonic components, that is, the value indicating the amount of echo leakage, falls within a certain range for limited purposes such as voice calls.

[0030] Therefore, the leakage coefficient used for calculating the amount of echo leakage is a constant, the amount of echo contained in the first signal is estimated from the leakage coefficient and the second signal, and the estimated value is The first signal force is subtracted or the ratio of the near-end signal included in the first signal is estimated from the leakage coefficient and the first and second signals, and this estimated ratio is used as the first signal. By multiplying, the first signal force can also remove the echo.

In the present invention, a constant is used as the leakage coefficient. For this reason, even when the near-end signal contains a large amount of noise, an echo having a large distortion caused by the echo path can be sufficiently suppressed. In addition, unlike the estimation of the amount of echo leakage shown in the third conventional example, no complicated calculation is required, so the calculation amount can be reduced. Therefore, echoes can be easily suppressed without being affected by near-end noise.

Brief Description of Drawings

FIG. 3 is a block diagram showing an example of the configuration of an echo suppressor of the present invention.

FIG. 4 is a block diagram showing an example of the configuration of the conversion unit shown in FIG.

[FIG. 5] FIG. 5 is a graph showing the experimental results of examining the correlation between the echo replica signal and the spectrum of the residual echo.

FIG. 6 is a schematic diagram showing a configuration example of a mobile phone device including a plurality of speakers and microphones. [FIG. 7] FIG. 7 is a graph showing the relationship between the leakage coefficient that can sufficiently suppress the echo and the power of the output signal of the linear echo canceller.

8] FIG. 8 is a block diagram showing the configuration of the first embodiment of the echo suppressor of the present invention.

FIG. 9 is a block diagram showing an example of the configuration of the coefficient generator shown in FIG.

FIG. 10 is a block diagram showing another configuration example of the coefficient generator shown in FIG.

FIG. 11 is a block diagram showing a configuration example of the spectral subtraction unit shown in FIG. 8.

[12] FIG. 12 is a block diagram showing a first configuration example of the Fourier coefficient subtracter shown in FIG.

13] FIG. 13 is a block diagram showing the configuration of the second embodiment of the echo suppressor of the present invention. FIG. 14 is a block diagram showing the configuration of the third embodiment of the echo suppressor of the present invention.

FIG. 15 is a block diagram showing a configuration example of a spectral subtraction unit shown in FIG.

FIG. 16 is a block diagram showing a first configuration example of the Fourier coefficient multiplier shown in FIG. 15.

FIG. 17 is a block diagram showing an example of the configuration of the smoothing section shown in FIG.

FIG. 18 is a block diagram showing another configuration example of the smoothing section shown in FIG.

FIG. 19 is a block diagram showing a second configuration example of the Fourier coefficient multiplier shown in FIG. 15.

FIG. 20 is a block diagram showing a third configuration example of the Fourier coefficient multiplier shown in FIG.

21] FIG. 21 is a block diagram showing the configuration of the fourth embodiment of the echo suppressor of the present invention. [22] FIG. 22 is a block diagram showing the configuration of the fifth embodiment of the echo suppressor of the present invention. FIG. 23 is a block diagram showing a configuration example of the echo canceller shown in FIG. FIG. 24 is a block diagram showing an example of the configuration of the spectral subtraction unit shown in FIG.

FIG. 25 is a block diagram showing the configuration of the sixth embodiment of the echo suppressor of the present invention. FIG. 26 is a block diagram showing the configuration of the seventh embodiment of the echo suppressor of the present invention. The best mode for carrying out the invention

Next, the present invention will be described with reference to the drawings.

FIG. 3 is a block diagram showing a configuration example of the echo suppression apparatus of the present invention.

As shown in FIG. 3, the echo suppressor of the present invention is a near-end signal generated by acoustic coupling of the microphone 1 and the speaker 2 in addition to the echo suppressor of the first conventional example shown in FIG. The coefficient generator 200 that generates a coefficient used to calculate the amount of leakage of the far-end signal (echo) that leaks into the filter (hereinafter referred to as the leakage coefficient), and the output signal of the microphone 1 or the output signal of the subtractor 4 When either one is the first signal and the output signal of the linear echo canceller 3 is the second signal, the first signal is generated based on the leakage coefficient generated by the coefficient generator 200 and the second signal. It further includes a conversion unit 100 that corrects the signal and outputs a near-end signal from which the first signal power echo is removed. The far-end signal input to speaker 2 is input from terminal 10, and the near-end signal is output from terminal 9.

Note that the linear echo canceller 3 may be a nonlinear echo canceller.

[0037] When the first signal and the second signal are divided into predetermined frequency regions, the coefficient generator 200 generates a leakage coefficient corresponding to the frequency region. At this time, the converter 100 corrects the first signal for each frequency domain using the corresponding leakage coefficient. Further, it is preferable that the coefficient generation unit 200 switches the leakage coefficient in accordance with a predetermined use situation set in advance.

FIG. 4 is a block diagram illustrating a configuration example of the conversion unit illustrated in FIG.

As shown in FIG. 4, the conversion unit 100 includes a frequency division unit 160, a frequency division unit 161, M correction units 166m (m = 1 to M), and a frequency synthesis unit 164.

[0040] The frequency divider 160 converts the first signal input via the terminal 162 into a predetermined frequency region. Divide into M for each area and output to the correction unit 166m corresponding to the frequency area. The frequency division unit 161 divides the second signal input via the terminal 163 into M for each predetermined frequency region, and outputs it to the correction unit 166m corresponding to the frequency region. The correction unit m corrects the first signal by using the leakage coefficient generated by the coefficient generation unit 200 and the second signal input via the terminal 167, and frequency-synthesizes the corrected signal. Output to part 164. The frequency synthesis unit 164 performs frequency synthesis on the output signal of the correction unit m and outputs the result from the terminal 165.

[0041] The correction unit 166m estimates the magnitude of the echo included in the first signal using the leakage coefficient and the second signal, and reduces the estimated echo magnitude by the first signal power. This corrects the first signal. The correction unit 166m estimates the ratio of the near-end signal included in the first signal based on the leakage coefficient, the first signal, and the second signal, and calculates the estimated ratio of the near-end signal to the first signal. The first signal may be corrected by multiplying by.

The frequency division units 160 and 161 perform frequency division using arbitrary linear transformation such as Fourier transform, cosine transform, subband analysis filter bank, and the like. The frequency synthesis unit 164 performs frequency synthesis using an inverse Fourier transform, an inverse cosine transform, a subband synthesis filter bank, or the like corresponding to the linear transformation used in the frequency division units 160 and 161.

[0043] The echo suppressor of the present invention differs from the third conventional example in that the amount of echo leakage is appropriately calculated from the first and second signals in that the leakage coefficient is a constant. In the third conventional example, since the amount of echo leakage depends on the frequency spectrum distribution of the far-end signal, it was recognized that it is inappropriate to set the leakage coefficient as a constant. However, the present inventor has confirmed through experiments that the echo can be sufficiently suppressed even if the leakage coefficient is a constant, insofar as the voice frequency spectrum distribution is different between women and men, as long as the purpose is voice communication. . Hereinafter, this point will be described in detail.

FIG. 5 is a graph showing the experimental results of examining the correlation between the echo replica signal and the spectrum of the residual echo. The horizontal axis of the graph shown in Fig. 5 shows the amplitude of the echo replica signal (the output amplitude of the linear echo canceller 3), and the vertical axis shows the amplitude of the residual echo (echo component included in the first signal). .

[0045] The correlation slope (residual echo amplitude Z echo replica amplitude) indicates the magnitude of the distortion of the echo, and the greater the slope, the greater the distortion. In other words, the slope of the correlation leaks out. It corresponds to the coefficient.

[0046] As shown in FIG. 5, it can be seen that the slope of the correlation varies depending on the frequency even for the same female voice. The same applies to men. However, when compared at the same frequency, the slope of the correlation between female voices and that of male voices are almost the same. As shown in Fig. 5, when the far-end signal is a sound whose spectral distribution is significantly different from human voice like music, even if it is the same frequency as the graph shown in Fig. 5 (1250Hz) 3125Hz), the slope of the correlation is completely different from human voice. The reason for this is that in music containing lower frequency components, there are far more frequency components that are the sources of harmonics that cause residual echo than human voices.

[0047] In this way, the slope of the correlation between the echo replica signal and the residual echo is dependent on the frequency vector distribution of the far-end signal. It was confirmed that they were similar. From this result, it can be seen that the same leakage coefficient may be used as long as the purpose is voice communication.

However, as shown in FIG. 5, the slope of the correlation between the echo replica signal and the residual echo differs depending on the frequency. Therefore, if the coefficient generator 200 generates a different leakage coefficient for each frequency domain of the first signal, and the converter 100 corrects the first signal using the leakage coefficient corresponding to the frequency domain, the echo is generated. Can be sufficiently suppressed.

[0049] By the way, the distortion sound of the echo said to be unable to be sufficiently suppressed by the linear echo canceller 3 is generated by the distortion sound generated by the speaker 2 itself and the vibration of the casing in which the microphone 1 and the speaker 2 are mounted. It is roughly divided into distorted sound. Furthermore, these distortions vary depending on the usage status of the device that is the object of echo suppression. Therefore, it is desirable that the coefficient generation unit 200 switches and outputs the leakage coefficient according to the usage status of the device that is the target of echo suppression.

[0050] Hereinafter, an example in which the leakage coefficient is switched in accordance with the use situation will be described using a cellular phone device as an example.

[0051] The cause of the distorted sound that also causes the speaker 2 itself is the nonlinearity of the speaker characteristics. Therefore, as shown in FIG. 6, in a mobile phone device that switches a plurality of speakers 301 to 303 as appropriate, if the characteristics of each speaker are different, the distortion of the echo depends on the speaker used. Is different. In such a usage situation, it is sufficient to detect the speaker to be used and switch the leakage coefficient according to the detected speaker power! / ヽ.

[0052] In addition, even in a mobile phone device in which only one speaker 2 is mounted, the amount of distorted sound reaching the microphone 1 from the speaker 2 changes depending on the positional relationship with the microphone 1, so that the distortion of the echo also changes. To do. In such a usage situation, the relative position between the speaker 2 and the microphone 1 may be detected, and the leakage coefficient may be switched according to the detected relative position. For example, in the case of the foldable mobile phone device 300 shown in FIG. 6, the positional relationship between the speaker 2 and the microphone 1 is determined by the angle of the hinge 321. Therefore, the angle of the hinge 321 is detected and leakage occurs according to the angle. What is necessary is just to switch a dust coefficient.

[0053] Also, in the folding mobile phone device 300 shown in Fig. 6, when the plurality of microphones 311 and 312 are switched as appropriate, the relative position with respect to the speaker 2 changes depending on the microphone used. In such a situation of use, it is only necessary to detect the microphone to be used and switch to a preset leakage coefficient according to the position of the detected microphone.

On the other hand, distorted sound resulting from the vibration of the casing is mainly generated at the joint between the components. For example, if the housing vibrates due to the output sound of the speaker 2 and a sound is generated in which the joint force between components is distorted, this distorted sound is input to the microphone 1 as an echo distortion. Therefore, when the volume of the speaker 2 changes, the acoustic energy transmitted from the speaker 2 to the housing changes, and the distorted sound generated at the joint between the parts also changes. In such a situation of use, it is only necessary to detect the volume setting value of the force 2 and switch the leakage coefficient according to the volume setting value.

[0055] Further, in the foldable mobile phone device 300 shown in FIG. 6, the amount of vibration of the housing changes depending on whether or not the force is completely folded, and the distortion sound generated at the joint between the parts also changes. Turn into. In such a use situation, it is only necessary to detect whether or not the cellular phone device 300 is completely folded and to switch the leakage coefficient according to the detection result.

[0056] Further, in the foldable mobile phone device 300 shown in FIG. 6, the position of the speaker changes depending on the bending angle, and therefore the sound transmitted from the speaker 2 depending on the angle of the hinge 321 even at the same part in the housing. Energy changes and occurs at the joint between parts Distorted sound changes. Therefore, even in such a usage situation, the angle of the hinge part 321 may be detected, and the leakage coefficient may be switched according to the angle.

[0057] Note that in a slide-type mobile phone device, the presence or absence of a slide and the amount of slide may be detected, and the leakage coefficient may be switched according to the detection result. In a mobile phone device equipped with both a slide mechanism and a folding mechanism, the angle of the hinge part, the force force force of the mobile phone device being folded, the presence or absence of a slide, or the amount of slide is detected, and leakage is detected according to the detection result. It is only necessary to switch the reconstitution coefficient. In mobile phone devices that are neither slide-type nor folding-type, for example, the factors that change the acoustic energy transmitted to the joints between components in the housing and the factors that affect the volume change of the echo are detected, and depending on the detection results It is only necessary to switch the leakage coefficient.

Furthermore, the present inventor has confirmed through experiments that the nonlinearity of the echo path changes as the power or amplitude of the signal output from the linear echo canceller 3 increases. In other words, when a distorted echo is generated in the state where the output signal of microphone 1 does not contain any near-end signal, the leakage coefficient that can sufficiently suppress the echo and the output signal of linear echo canceller 3 Examining the relationship with power, the results shown in Fig. 7 were obtained. FIG. 7 shows the relationship between the output signal of the linear echo canceller 3 in the frequency band centered at 1875 Hz and the leakage coefficient corresponding thereto. The horizontal axis of the graph shown in Fig. 7 shows the power of the output signal of the linear echo canceller 3, and the vertical axis shows the leakage coefficient that can sufficiently suppress the echo.

[0059] The leakage coefficient that can sufficiently suppress the echo changes abruptly when the power value of the output signal of the linear echo canceller 3 reaches 2000000 as shown in FIG. This is because when the power of the output signal of the linear echo canceller 3 is large, the power of the input signal of the linear echo canceller 3, that is, the far-end signal input to the speaker 2, is also large. This is thought to be due to a sharp increase in strain.

Therefore, in the echo suppression device of the present invention, the power or amplitude of the signal output from the linear echo canceller 3 is detected as the usage status, and the leakage coefficient is switched according to the detected value. Such a method uses the power and vibration of the output signal of the linear echo canceller 3. Instead of the width, it is also possible to use the power and amplitude of the far-end signal, or the power and amplitude of a specific frequency component included in the far-end signal.

The method for switching the leakage coefficient based on the output signal of the linear echo canceller 3 is similar to the method for switching the leakage coefficient based on the sound volume 2 setting value. However, since the latter has no far-end signal, the leakage coefficient is selected according to the volume setting even when echo suppression is not required. On the other hand, the former is superior in that it does not select such a leakage coefficient by mistake.

[0062] The method for switching the leakage coefficient described above does not need to detect all of the above-mentioned usage conditions and switch the leakage coefficient, and detects one or more of the usage conditions, and sets the leakage coefficient. You may switch.

[0063] For example, in a situation where a mobile phone device equipped with a plurality of cameras is used to make a call while exchanging images with each other, (according to a so-called videophone), depending on the camera used by the mobile phone device If the microphone and speaker are automatically switched, you can use the image information captured by the camera to detect the V or microphone and speaker instead of directly detecting the microphone mouthphone or speaker to be used! /.

[0064] When the usage situation used for switching the leakage coefficient is determined, the optimum leakage coefficient corresponding to the usage situation is determined by experiment or computer simulation, and the coefficient is generated by associating the leakage coefficient with the usage situation. Stored in part 200.

It should be noted that the use state that can be detected by a sensor provided outside the echo suppression device, such as the angle of the hinge part, the speaker volume setting value, the speaker to be used, etc., is input to the coefficient generator 200. That's fine. On the other hand, the usage of the power and amplitude of the far-end signal, the power and amplitude of the output signal of the linear echo canceller 3, and the power and amplitude of specific frequency components included in the far-end signal are detected in the echo suppressor, The detection result may be input to the coefficient generator 200.

[0066] According to the echo suppression apparatus of the present invention, by setting the leak coefficient as a constant, the leak coefficient that is a constant is not affected by noise, and therefore, even in an environment where large noise is input as near-end speech, It is possible to sufficiently suppress the echo generated due to. In addition, since complicated calculations such as the estimation of the amount of echo leakage shown in the third conventional example are not required, The amount can be reduced. Therefore, echo can be easily suppressed without being affected by near-end noise.

[0067] In particular, by selecting the optimum leakage coefficient according to the use situation, even an echo generated due to distorted sound can be satisfactorily suppressed.

Next, an embodiment of the echo suppression device of the present invention will be described with reference to the drawings.

[First Example]

FIG. 8 is a block diagram showing the configuration of the first embodiment of the echo suppressor of the present invention.

[0069] The echo suppression apparatus of the first embodiment is an example in which the spectral sub-translation unit 6 is used as the conversion unit 100 shown in FIG.

The coefficient generation unit 200 of the first embodiment generates a leakage coefficient indicating the amount of echo leakage generated by acoustic coupling between the microphone 1 and the speaker 2 as described above.

[0071] The spectral subtraction unit 6 receives the output signal of the subtractor 4, the output signal of the linear echo canceller 3, the leakage coefficient generated by the coefficient generation unit 200, and the voice detection result of the voice detection unit 5. The

[0072] The spectral subtraction unit 6 divides the output signal of the subtractor 4 and the output signal of the linear echo canceller 3 into predetermined frequency regions, respectively, and removes echoes for each signal component in the decomposed frequency region. To do.

A coefficient generation unit 200 shown in FIG. 9 is configured to include a coefficient storage unit 201 that holds a leakage coefficient suitable for each frequency region from band 1 to band M.

The coefficient generation unit 200 reads out the leakage coefficient for each frequency region (band) stored in the coefficient storage unit 201 and outputs it to the spectral subtraction unit 6. These leakage coefficients correspond to, for example, the correlation slope at a frequency of 1250 Hz and the correlation slope at a frequency of 3125 Hz shown in FIG.

A coefficient generation unit 200 shown in FIG. 10 includes a coefficient storage unit 202 that holds a leakage coefficient group suitable for each frequency region from band 1 to band M, and a system including the echo suppression device of the present invention. And a usage status detection unit 203 for detecting various usage statuses of the system.

[0077] Coefficient generation section 200 shown in FIG. 10 includes a coefficient storage section that stores a leakage coefficient corresponding to the usage status detected by usage status detection section 203 among the leakage coefficient group corresponding to each frequency domain.

Read from 202 and output to spectral subtraction unit 6.

[0078] In the configuration shown in FIG. 10, the leakage coefficient group corresponding to each frequency region includes the leakage coefficient for usage condition 1, the leakage coefficient for usage condition 2, and so on. It has a leakage coefficient. N is an arbitrary value of 2 or more.

[0079] For example, as an example of the usage situation, when detecting the volume setting value of the speaker 2, the usage situation detection unit 203 detects a volume setting value of the speaker 2, a detected volume setting value, and a predetermined value. A discriminator for comparing the threshold value and converting the comparison result into a digital value of 2 or more.

[0080] As another example of the usage situation, there is a method of detecting the angle of the hinge portion in the folding mobile phone device. In this case, the usage status detection unit 203 compares a sensor (not shown) that detects the angle of the hinge unit with the detection angle and a predetermined threshold value, and converts the comparison result into a digital value of two or more values. (Not shown).

[0081] As another example of the usage situation, when detecting a speaker that is in use as a mobile phone device equipped with a plurality of speakers, the usage situation detector 203 determines which speaker is being used, and determines It has a decision unit (not shown) that outputs the result as a digital value of two or more values.

[0082] As another example of the usage situation, when detecting a microphone being used from a mobile phone device having a plurality of microphones, the usage situation detection unit 203 determines which microphone is being used, and determines It has a judgment unit (not shown) that outputs the result as a digital value of two or more values.

[0083] As another example of the usage situation, when detecting the power or amplitude of the output signal of the linear echo canceller 3, the usage situation detection unit 203 detects the power or amplitude of the output signal of the linear echo canceller 3. (Not shown) and a discriminator (not shown) for judging the detected power or amplitude as a threshold value and converting it into a digital value of two or more values. For example, if the system including the echo suppressor of the present invention has the characteristics shown in the graph of FIG. 5, the required leakage coefficient is 1 when the output power of the linear echo canceller 3 is 2000000. Therefore, if the threshold value is set to 2000000, “0” is output if it is less than 2000000, and “1” is output if it exceeds 2000000!

[0084] In addition, any use condition that affects the amount of echo leakage can be used. It is also possible to use a plurality of usage conditions in combination.

[0085] The coefficient storage unit 202 selects one corresponding to the output signal of the usage status detection unit 203 from a plurality of leakage coefficients registered in advance corresponding to each frequency region, and selects the selected leakage factor. The reconstitution coefficient is output to the spectral subtraction unit 6.

[0086] For example, when the power characteristic of the output signal of the linear echo canceller 3 shown in FIG. 7 is used as a usage situation, “1” shown by the thick solid line in FIG. 7 corresponding to the frequency region centered on 1875 Hz. ”And“ 20 ”are retained, and if“ 0 ”is output from the usage detection unit 203,“ 1 ”is output as the leakage coefficient, and“ 1 ”is output from the usage detection unit 203. When is output, “20” is output as the leakage coefficient.

Spectral subtraction section 6>

FIG. 11 is a block diagram showing a configuration example of the spectral subtraction unit shown in FIG.

As shown in FIG. 11, the spectral subtraction unit 6 has a configuration including a Fourier transform 60, a Fourier transform 61, a Fourier coefficient subtractor 66m (m = 1 to M), and an inverse Fourier shelf 64. is there.

[0088] The Fourier transformer 60 performs an M-point Fourier transform process on the output signal of the subtractor 4, and uses the processing result (amplitude and phase) as the first Fourier coefficient to apply a Fourier coefficient corresponding to each frequency domain. Output to subtractor 66m (m = 1 to M).

[0089] The Fourier transform 61 performs M-point Fourier transform processing on the echo replica signal output from the linear echo canceller 3, and the processing result (amplitude and phase) corresponds to each frequency domain as the second Fourier coefficient. Output to the Fourier coefficient subtractor 66m.

[0090] The Fourier coefficient subtractor 66m outputs the first Fourier coefficient output from the Fourier transformer 60, the second Fourier coefficient output from the Fourier transform 61, and the coefficient generator 200 shown in FIG. The leakage coefficient is received, and the Fourier coefficient is calculated by performing the subtraction process using those amplitude components, and the calculation result (amplitude and phase) is inverse Fourier transformed. Output to device 64.

The inverse Fourier transform ^^ 64 performs an inverse Fourier transform process on the Fourier coefficient group output from the Fourier coefficient subtraction units 661 to 66M, and outputs a real part of the processing result.

Next, the Fourier coefficient subtractor 66m (m = 1 to M) shown in FIG. 11 will be described with reference to FIG.

FIG. 12 is a block diagram showing a first configuration example of the Fourier coefficient subtracter shown in FIG.

The first Fourier coefficient for each frequency domain output from Fourier transformer 60 shown in FIG. 11 is supplied to subtractor 706 via terminal 700.

The second Fourier coefficient output from the Fourier transform shown in FIG. 11 is supplied to a multiplier 707 via a terminal 703. Further, the leakage coefficient generated by the coefficient generator 20 is supplied to a multiplier 707 via a terminal 167.

[0096] Multiplier 707 multiplies the leakage coefficient by the second Fourier coefficient, and subtracts the multiplication result.

Output to 706. The subtractor 706 also subtracts the output value of the multiplier 707 from the first Fourier coefficient force.

, The calculation result is output. The calculation result of the subtractor 706 is output to the inverse Fourier transformer 64 shown in FIG.

[0097] Here, the multiplier 707 multiplies the leakage coefficient and the second signal coefficient calculated from the output signal power of the linear echo canceller 3 by the multiplier 707, and the multiplier 707 converts the first Fourier coefficient to An estimate of the Fourier coefficient from the remaining echo is obtained. By subtracting the first Fourier coefficient force by the subtractor 706, the estimated Fourier coefficient value of the near-end signal with the echo component suppressed is obtained.

The estimated value for each frequency domain is synthesized by inverse Fourier transformation 64 shown in FIG. 11, and is output as a near-end signal. As a result, the combined near-end signal is a signal with echo suppressed.

The operation of the Fourier coefficient subtractor 66m described above will be described using equations.

[0100] If the Fourier coefficient of the near-end signal is S, the near-end speech component included in the near-end signal is A, the echo component is E, and the noise component is N,

S = A + E + N

There is a relationship. [0101] Also, let R be the Fourier coefficient of the echo replica signal and P1 be the value of the leakage coefficient. P 1 is an approximate value of the rate at which the far-end signal R leaks into the near-end signal as an echo, and corresponds to the echo gain in the echo path. Incidentally, in the echo suppressor of the third conventional example, P1 is expressed by the following equation.

[0102] PI = Av [S / R] = Av [(E + N) / R] · '· (2)

Here, Αν [·] represents a smoothing process.

Therefore, a value Ρ2 (corresponding to the output signal of the multiplier 707) obtained by multiplying the leakage coefficient P1 by the Fourier coefficient R of the echo replica signal is an estimated value of the echo component.

P2 = P1 XR

= Ex [E]--(3)

Here, Εχ [·] indicates the estimated value.

[0104] The value obtained by subtracting Ρ2 from S Ρ3 (equivalent to the output signal of subtractor 706: near-end signal) is

P3 = S -P2

= S— P1 XR

= A + E + N-Ex [E]

It becomes. That is, the output of the subtractor 706 is an estimated value of the sum of the Fourier coefficient component Α and the noise component の of the near-end speech from which the echo component Ε has been removed.

Next, how the echo suppression apparatus of the first embodiment shown in FIG. 8 operates when distortion occurs in a speaker or the like in the echo path will be described.

When distortion occurs in the echo path, the echo suppression apparatus of the first embodiment removes distortion components in the echo by non-linear calculation in the frequency domain in the spectrum subtraction unit 6. The echo suppressor of the first embodiment effectively removes distortion components contained in echoes by adjusting the time variation of signal components important in frequency domain nonlinear calculations by the linear echo canceller 3.

[0107] The output signal of the microphone 1 includes an echo generated due to distortion of the far-end signal in addition to the far-end signal. This echo can be thought of as the harmonic component of the far-end signal. Hereinafter, in order to simplify the explanation, consider a case where the echo component E is only a harmonic component due to distortion.

[0109] As can be seen from the above equation (3), the spectral subtraction unit 6 can remove the echo component E in principle unless the far-end signal has a Fourier transform coefficient R of zero. It is possible. Here, what is important for removing the echo component E is the accuracy of the leakage coefficient P 1 corresponding to the gain of the echo in the echo path.

[0110] In the echo suppressor of the third conventional example, the amount of echo leakage is estimated when the near-end speech is not detected for the microphone output signal power based on the speech detection result. Therefore, it is difficult to accurately detect the voice. If there is an error in the voice detection result, the leakage coefficient P1 becomes an abnormally large value, and the near-end signal P3 calculated based on the incorrect leakage coefficient P1 also deteriorates. That is, the echo included in the near-end signal P3 is not sufficiently suppressed, and a large distortion occurs in the near-end speech. In order to avoid such problems, if the control is performed so that the leakage coefficient P1 is not updated, if the echo leakage amount fluctuates, the error of the leakage coefficient P1 increases, and the calculation is based on the leakage coefficient P1. The near-end signal P3 is also degraded.

[0111] For example, when the hands-free telephone is a foldable mobile phone device or when the speaker to be used can be switched, the amount of echo leakage depends on the angle of the folding hinge, the force used, etc. Fluctuates. In an environment where there is a lot of near-end noise, it is common for speakers to change the usage status, such as changing the angle of the hinge or switching the speaker to be used so that the sound can be heard well. In that case, the corrected signal of the first signal will deteriorate.

[0112] On the other hand, in the present embodiment, a constant set in advance according to the state of use is used as the leakage coefficient P1. Therefore, if the angle of the hinge part and the speaker to be used are detected, the leakage coefficient P1 can be obtained without being affected by the noise contained in the near-end signal.

[0113] According to an experiment conducted by the present inventor using a mobile phone device, rather than using an estimated value with a large error as the leakage coefficient P1, a constant set in advance according to the use situation is used as the leakage coefficient P1. It is better to use this method for removing echo and near-end speech distortion. [0114] Further, the echo suppression apparatus of the first embodiment has an effect of removing residual echoes even when the linear echo canceller 3 shown in Fig. 8 performs erroneous echo path estimation.

[0115] In the above description, the case where the echo component E is only the harmonic component due to distortion is considered. The echo component of the far-end signal not caused by force distortion, that is, the echo component excluding the harmonic component is also described in this embodiment. This echo suppressor can suppress.

[0116] For example, if the echo path estimation is wrong in the linear echo canceller 3, the subtracter 4 shown in FIG. However, even in such a case, since the far-end signal component is removed by the spectral subtraction unit 6, the echo is sufficiently suppressed.

[0117] Also, the echo suppression apparatus of the present embodiment has an echo suppression effect by the spectral subtraction unit 6, thereby reducing the number of taps of the linear echo canceller 3 (the number of taps of the adaptive filter). The amount can be reduced.

[0118] Since the echo suppressor of the first conventional example shown in FIG. 1 has only the linear echo canceller 3, the echo removal capability can be achieved by reducing the number of taps of the adaptive filter provided in the linear echo canceller 3. Is reduced. However, in the echo suppressor of the first embodiment shown in FIG. 8, the provision of the spectral subtraction unit 6 compensates for the reduction of the echo removal capability even if the number of taps of the adaptive filter is reduced. Therefore, an echo suppression device having sufficient echo removal capability can be obtained.

[0119] The echo suppressor of the first embodiment includes the linear echo canceller 3 and the frequency domain nonlinear calculation by the spectrum subtraction unit 6, and sufficient echo removal is achieved by compensating each other's poor processing. Get the ability.

[0120] That is, when the echo path is distorted or when the echo path estimation is wrong with the linear echo canceller 3, the echo cannot be sufficiently suppressed with the linear echo canceller 3 alone. Echo can be suppressed.

[0121] Further, by correcting the output signal of the microphone using the output signal of the linear echo canceller 3, it is possible to take into account the time lag that cannot be dealt with only by the frequency domain nonlinear calculation by the spectral subtraction unit 6. Harmonic components that cause distortion can be suppressed by a simple estimation process using only amplitude values. [0122] In addition, by using a preset constant for the leakage coefficient PI used in the spectral subtraction unit 6 according to the usage status, for example, even when the usage status is changed in an environment where the near-end noise is high, The near-end speech with less distortion can be obtained by suppressing the echo sufficiently.

[0123] Furthermore, unlike the echo suppressor of the third conventional example, the echo suppressor of the first embodiment does not require a complicated calculation process for estimating the amount of leakage of the echo. The amount is reduced.

[Second Example]

FIG. 13 is a block diagram showing the configuration of the second embodiment of the echo suppressor of the present invention.

The echo suppressor of the second embodiment is different from the echo suppressor of the first embodiment in that the output signal of the microphone 1 is input to the spectral subtraction unit 6 instead of the output signal of the subtractor 4. .

[0125] In the echo suppressor of the first embodiment, the power of removing the main component of the echo by the linear echo canceller 3 In the echo suppressor of the second embodiment, the main component of the echo is removed by the spectral subtraction unit 6. To do. Other configurations and operations are the same as in the first embodiment, and the effect of removing echoes caused by distortion can be obtained in the same manner as in the first embodiment.

Therefore, the echo suppressor of the second embodiment also has a linear echo canceller as in the first embodiment, such as when the acoustic transmission system is distorted or when the echo path estimation is erroneous in the linear echo canceller 3. Even if the echo cannot be sufficiently suppressed by 3 alone, the echo can be sufficiently suppressed by the spectral subtraction unit 6.

[0127] Further, by using a constant preset according to the use situation as the leakage coefficient P1 used in the spectral subtraction section 6, even if the use situation is changed in an environment where the near-end noise is high, the echo It is possible to obtain near-end speech with little distortion.

[0128] In addition to the configurations shown in the first and second embodiments, the spectral subtraction unit 6 is, for example, non-patent document 2 (Xiao jian Lu, Benoit Champagne's statement "Acousal Echo Cancellation Over A Non-Linear Channel ", International Workshop on Spectral Subtraction described in Acoustic Echo and Noise Control 2001), or Non-Patent Document 3 (A. Alvarez et al. "A Speech

It is also possible to use the spectral subtraction described in “Enhancement Based On Negative Beamrorming And Spectral Bubtration”, International Workshop on Acoustic Echo and Noise Control 2001).

[Third embodiment]

FIG. 14 is a block diagram showing the configuration of the third embodiment of the echo suppressor of the present invention.

[0129] The echo suppression apparatus of the third embodiment is different from the echo suppression apparatus of the first embodiment in that a spectrum subtraction section 7 is used instead of the spectrum subtraction section 6 shown in FIG. . Since other configurations and operations are the same as those in the first embodiment, a detailed description thereof will be omitted.

Hereinafter, the spectral substituting unit 7 shown in FIG. 14 will be described with reference to the drawings.

FIG. 15 is a block diagram showing an example of the configuration of the spectral subtraction unit shown in FIG.

[0132] As shown in FIG. 15, the spectral substituting unit 7 includes a Fourier transformer 70, a Fourier transform 71, a Fourier coefficient multiplier 76m (m = l to M), and an inverse feeler transform 74. It is.

[0133] The Fourier transformer 70 performs M-point Fourier transform processing on the output signal of the subtractor 4 shown in Fig. 14 that is input via the terminal 72, and outputs the processing result (amplitude and phase). 1 Fourier coefficient is output to the Fourier coefficient multiplier 76m (m = l to M) corresponding to each frequency domain.

The Fourier transformer 71 performs M-point Fourier transform processing on the output signal (echo replica signal) of the linear echo canceller 3 shown in FIG. Amplitude and phase) are output as second Fourier coefficients to Fourier coefficient multiplier 76m corresponding to each frequency domain.

[0135] The Fourier coefficient multiplier 76m receives the first Fourier coefficient output from the Fourier transform 70, the second Fourier coefficient output from the Fourier transform 71, and the terminal 67. 14 receives the leakage coefficient output from the coefficient generator 200 shown in FIG. 14 and performs a multiplication process using these amplitude components to calculate the Fourier coefficient, and the calculation result (amplitude and phase) Is output to the inverse Fourier transform 74.

[0136] Inverse Fourier transform 74 performs an inverse Fourier transform process on the family coefficient group output from Fourier coefficient multiplier 76m (m = l to M), and outputs the real part of the processing result from terminal 75. To do. In the configuration shown in FIG. 15, a near-end signal in which the echo component is suppressed by the Fourier coefficient multiplier 76m (m = 1 to M) is obtained.

Next, the configuration and operation of the Fourier coefficient multiplier 76m (m = 1 to M) will be described with reference to FIG.

FIG. 16 is a block diagram showing a first configuration example of the Fourier coefficient multiplier shown in FIG.

As shown in FIG. 16, the Fourier coefficient multiplier of the first configuration example includes an absolute value calculation unit 731, an absolute value calculation unit 734, a multiplier 737, a divider 745, a multiplier 746, and a smoothing unit 747. And a subtractor 744.

The first Fourier coefficient for each frequency domain output from the Fourier transform shown in FIG. 15 is output to absolute value calculating section 731 and multiplier 737 via terminal 730. The second Fourier coefficient output from the Fourier transform shown in FIG. 15 is output to the absolute value calculation unit 734 via the terminal 733.

[0141] Absolute value calculation unit 731 calculates the absolute value of the first Fourier coefficient, and outputs the calculation result to divider 745. Absolute value calculation section 734 calculates the absolute value of the second Fourier coefficient, and outputs the calculation result to divider 745. Divider 745 divides the calculation result of absolute value calculation unit 734 by the calculation result of absolute value calculation unit 731 and outputs the calculation result to multiplier 746.

Multiplier 746 multiplies the leakage coefficient generated by coefficient generator 200 input from terminal 167 and the output signal of divider 745, and outputs the calculation result to smoother 747. The smoothing unit 747 smoothes the output signal of the multiplier 746 and outputs it to the subtracter 744.

The subtractor 744 subtracts the output value of the smoothing unit 747 from the value “1.0”, and outputs the calculation result to the multiplier 737. Multiplier 737 multiplies the output value of subtractor 744 and the first Fourier coefficient output from Fourier transform 70, and outputs the multiplication result. Output signal of multiplier 737 The signal is output to the inverse Fourier transform 74 shown in FIG.

FIG. 17 is a block diagram showing a configuration example of the smoothing unit shown in FIG.

A smoothing unit 747 shown in FIG. 17 includes a subtracter 801, a multiplier 802, an adder 803, a limiter 807, and a delay 804.

The input signal of the smoothing unit 747 (the output signal of the multiplier 746) is supplied to the subtracter 801 via the terminal 800. The subtractor 801 is a delay unit 8 that delays the input signal by one sample time.

The output signal of 04 (the output signal of the smoothing unit) is subtracted and the calculation result is output to the multiplier 802.

Multiplier 802 multiplies the output signal of subtractor 801 and the smoothing coefficient input via terminal 806, and outputs the calculation result to adder 803. Adder 803 adds the output signal of multiplier 802 and the output signal of delay unit 804, and outputs the calculation result to limiter 807. The limiter 807 limits the amplitude of the output signal of the adder 803 within a predetermined upper limit value and lower limit value, and outputs the limited signal to the output terminal 899 and the delay device 804. The delay device 804 delays the output signal of the limiter 807 by one sample time, and outputs the delayed signal to the subtractor 8001 and the adder 803.

The smoothing unit 747 shown in FIG. 17 has a configuration called a so-called leak integrator or first-order IIR type low-pass filter. In the smoothing unit 747 shown in FIG. 17, the input smoothing coefficient and the time constant of the smoothing process are in an inversely proportional relationship. The smoothing unit 747 may adopt an arbitrary configuration having a smoothing effect such as a high-order IIR filter, which is not limited to the configuration shown in FIG.

A smoothing unit 747 shown in FIG. 18 has a configuration including a smoothing coefficient determination unit 810 that generates a smoothing coefficient in addition to the smoothing unit shown in FIG. The smoothing coefficient determination unit 810 also generates a smoothing coefficient for the output signal power of the subtractor 801, and outputs the smoothing coefficient to the multiplier 802. In such a configuration, the rising speed and falling speed of the output signal of the smoothing unit 747 can be set to different values.

[0151] The smoothing coefficient determination unit 810 outputs a relatively small coefficient, for example, 0.001, when the output signal of the subtractor 801 is positive, that is, when the output signal of the subtractor 801 increases, and performs subtraction. When the output value of the subtracter 801 is negative, that is, when the output signal of the subtractor 801 decreases Outputs a relatively large coefficient, eg, 0.01.

[0152] When the smoothing coefficient is set in this way, the speed at which the output signal of the smoothing section 747 increases, that is, the rising speed becomes slow, and the speed at which the output signal of the smoothing section 747 decreases, that is, the falling speed, Get faster. Therefore, the output speed of the subtracter 744 shown in FIG. 16, that is, the estimated speed of the near-end speech and near-end noise included in the near-end signal, increases at the rising speed, and the falling speed. Becomes slower.

[0153] In general, the amplitude change of voice or music, that is, the envelope characteristic, is often fast when rising and slow when falling. The smoothing section shown in FIG. 18 can have such envelope characteristics, and can improve the estimation accuracy of the ratio of near-end speech and near-end noise included in the near-end signal.

Here, the operation of subtracter 744 shown in FIG. 16 will be described using mathematical expressions.

[0155] The value P4 smoothed by dividing the second row of the above equation (4) by S can be expressed by the following equation (5). The right side of equation (5) corresponds to the output value of the subtracter 744 shown in FIG.

P4 = Av [P3 / S]

= Av [l-{(R / S) XAv [(E + N) / R]}]

= l -Av [{(R / S) XAv [(E + N) / R]}] · '· (5)

The value P4 is the smoothed value obtained by dividing the third row of equation (4) by S.

P4 = Av [{(A + E + N) -Ex [E + N]} / S]

= Av [Ex [A] / S]

= Ex [A / S]--(6)

Can also be expressed. Comparing equation (6) and equation (5), it can be divided that the output value P4 of the subtracter 744 is an estimated value of the ratio of the near-end speech included in the near-end signal.

Therefore, by using the multiplier 737 shown in FIG. 16 and multiplying the output value of the subtractor 744 by the output signal of the subtracter 4 shown in FIG. 14, the signals other than the echoes included in the near-end signal can be obtained. An estimate of the Fourier coefficient of the near-end speech with suppressed signal, ie echo, is obtained. By combining the estimated values with the inverse Fourier transform shown in Fig. 15, a near-end signal with suppressed echo is obtained.

Next, when distortion occurs in the speaker 2 etc. in the echo path, the echo suppression device of the third embodiment is used. We will explain how the device works.

[0158] As shown in equations (5) and (6), the output value P4 of the subtractor 744 shown in Fig. 16 is an estimated value of the proportion of the near-end speech included in the near-end signal.

[0159] This value P4 is calculated using P3 shown in the first embodiment as shown in the equation (5). As described in the first embodiment, P3 is an estimated value of the Fourier coefficient component of the near-end speech, and not only the echo component and noise component, but also the echo of the harmonic component generated by the distortion is removed. Therefore, P4 is also a value obtained by removing the echo of the harmonic component generated by the distortion, and the distortion echo component is also suppressed in the Fourier coefficient obtained by multiplying this P4.

[0160] As described above, the output signal of microphone 1 includes not only the far-end signal (echo component).

Also included are echoes due to far-end signal distortion. The echo due to this distortion can be considered as the harmonic component of the far-end signal.

[0161] According to the echo suppression apparatus of the third embodiment, by including the spectrum substituting unit 7,

By using harmonic components contained in the far-end signal, it is possible to suppress echoes generated by distortion of the far-end signal.

[0162] That is, in the echo suppressor of the third embodiment as well, when the echo path is distorted or when the echo path estimation is incorrect in the linear echo canceller 3, the echo is sufficiently suppressed by the linear echo canceller 3 alone. Even if this is not possible, the spectral subtraction unit 7 can sufficiently suppress the echo.

[0163] Furthermore, by using a constant set in advance according to the use situation as the leakage coefficient, even if the use situation is changed in an environment where there is a large amount of near-end noise, for example, the echo is sufficiently suppressed and distorted. You can get near-end audio!

FIG. 19 is a block diagram showing a second configuration example of the Fourier coefficient multiplier shown in FIG.

[0165] The Fourier coefficient multiplier 76m in the second configuration example includes the smoothing unit 740 inserted in the signal path from the absolute value calculation unit 731 to the divider 745, and the signal path from the absolute value calculation unit 734 to the divider 745. 16 differs from the first configuration example shown in FIG. 16 in that a smoothing portion 741 is inserted.

[0166] The smoothing unit 740 and the smoothing unit 741 may have the same configuration as the smoothing unit 747 except that the smoothing coefficients are different. Therefore, detailed description thereof is omitted here. In the Fourier coefficient multiplier 76m shown in FIG. 19, since the input value of the divider 745 is smoothed by the smoothing unit 740 and the smoothing unit 741, the smoothing unit 747 passes from the divider 745 through the multiplier 746. The value supplied to is also smoothed. Therefore, a more stable output value can be obtained from the smoothing unit 747 than the Fourier coefficient multiplier 76m of the first configuration example shown in FIG.

[0168] The Fourier coefficient multiplier 76m in the first configuration example and the second configuration example is the configuration in which the estimated value of the ratio of the near-end speech included in the near-end signal is obtained from the subtractor 744.

Accordingly, when the Fourier coefficient multiplier 76m of the second configuration example shown in FIG. 19 is used, the same as the case of using the Fourier coefficient multiplier 76m of the first configuration example shown in FIG. The effect of the present invention can be obtained.

[0171] The Fourier coefficient multiplier 76m of the third configuration example is different from the second configuration example shown in Fig. 19 in that the processing order of the paths from the smoothing unit 740 and the smoothing unit 741 to the multiplier 737 is different.

[0172] In the Fourier coefficient multiplier 76m of the third configuration example, the output value of the smoothing unit 740 is output to the subtractor 744 and the divider 745, and the output value of the smoothing unit 741 is output to the multiplier 746.

Multiplier 746 multiplies the output value of smoothing section 741 by the leakage coefficient generated by coefficient generation section 200 and outputs the calculation result to subtractor 744. The subtracter 744 also subtracts the output value of the multiplier 746 from the output value of the smoothing unit 740 and outputs the calculation result to the divider 745. Divider 745 divides the output value of subtractor 744 by the output value of smoothing unit 740 and outputs the calculation result to smoothing unit 748. Smoothing section 748 smoothes the output value of divider 745 and outputs the processing result to multiplier 737.

[0174] The smoothing unit 748 may have the same configuration as that of the smoothing unit 747 except that the smoothing coefficients are different.

Here, if the configuration shown in FIG. 18 is adopted for smoothing section 748, it is possible to have a slow envelope characteristic at the time of falling fast and at the time of falling, and the near-end speech included in the near-end signal In addition, the estimation accuracy of the ratio of near-end noise can be improved.

[0176] The output value P5 of the smoothing unit 748 is expressed by the following equation (7).

P5 = Av [(Av [S] —PI X Av [R] (S) / Av [S])] = Av [(Av [((A + Ε + Ν) S— Εχ [Ε])) / Av [S]]

= Ex [(A + N) / S]-(7)

From Expression (7), it can be seen that the output value P5 of the smoothing unit 748 is an estimated value of the ratio of the near-end speech included in the near-end signal, as in the case of P4.

Accordingly, the Fourier coefficient multiplier 76m of the third configuration example shown in FIG. 20 also has the same function as the second configuration example shown in FIG. 19, and the Fourier coefficient of the first configuration example shown in FIG. Similar to the case of using the multiplier 76m, the above-described effects of the present invention can be obtained.

[Fourth embodiment]

FIG. 21 is a block diagram showing the configuration of the fourth embodiment of the echo suppressor of the present invention.

[0178] The echo suppression apparatus of the fourth embodiment is the same as that of the third embodiment shown in FIG. 14 in that the output signal of the microphone 1 is input to the spectrum substituting unit 7 instead of the output signal of the subtractor 4. Unlike the co-suppressor.

Therefore, in the echo suppressor of the third embodiment, the echo main component is removed by the linear echo canceller 3, whereas in the echo suppressor of the fourth embodiment, the echo is suppressed by the spectrum sub-recession unit 7. Remove major components.

[0180] Other configurations and operations are the same as those of the third embodiment, and the effect of removing echoes caused by distortion can be obtained in the same manner as in the third embodiment.

Therefore, the echo suppressor of the fourth embodiment also has a linear echo canceller as in the third embodiment, such as when the acoustic transmission system is distorted or when the echo path estimation is wrong in the linear echo canceller 3. Even if only 3 cannot suppress the echo sufficiently, the spectrum suppression unit 7 can sufficiently suppress the echo.

[0182] Furthermore, by using a value set in advance according to the usage status as the leakage coefficient used in the spectrum sub-restriction unit 7, even if the usage status is changed in an environment where the near-end noise is high, the echo is used. It is possible to obtain near-end speech with little distortion.

Although the embodiments of the present invention have been described above, the present invention is not limited to the first to fourth embodiments described above, and various modifications as described below are possible. For example, in the first conventional example to the fourth conventional example, the spectral subtraction unit 6 and the spectral subtraction unit 7 have been described as examples in which Fourier transform is performed at predetermined sample periods. It is possible to process in units of frames at regular intervals, not limited to each cycle.

[0185] Also, it is possible to process by overlapping frames. At that time, it is possible to reduce the amount of computation by using techniques such as overlap wrap save and overlap add. For more information on overlap save and overlap add, see Non-Patent Document 4 (the paper requency-Domain and Multirate Adaptive Pi Itering by John J. Shynk, IEEE Signal Processing Magazine, January 1992, pp. 14-37). Are listed.

[0186] In the first conventional example to the fourth conventional example, in addition to the force Fourier transform described in the example in which Fourier transform is performed in the spectrum subtraction unit 6 and the spectrum subthreshold unit 7, linear such as cosine transform and filter bank is used. Conversion can also be used, and it is also possible to perform processing after conversion to the subband region. In that case, the Fourier coefficient subtractor and multiplier may be changed in accordance with their linear transformation. For example, when using cosine transform, use a subtractor for cosine coefficients and a multiplier for cosine coefficients. The operation of these various arithmetic units is the same as that in the case where the Fourier transform is used for the linear transformation shown in the first to fourth conventional examples.

[Fifth embodiment]

In the first to fourth embodiments, the example using the linear echo canceller 3 has been described. However, it is also possible to use a transform domain echo canceller for echo suppression. In that case, if the conversion region of the conversion region echo canceller is the same conversion region as the spectral subtraction unit 6 and the spectral subtraction unit 7 described above, the amount of calculation of the entire echo suppression device is reduced and the delay time associated with the calculation is reduced. Can be shortened.

[0187] Note that the transform domain echo canceller is an echo canceller that performs echo suppression processing on the transform domain expanded by linear transform and re-synthesizes the original domain by inverse linear transform.

[0188] Hereinafter, as a transform domain echo canceller, for example, a buffer described in Non-Patent Document 4 is used. A description will be given using an example in which a one-lier transform area echo canceller is used.

FIG. 22 is a block diagram showing the configuration of the fifth embodiment of the echo suppressor of the present invention.

The echo suppression apparatus of the fifth embodiment has a configuration in which the echo canceller 13 and the spectral subtraction unit 16 perform processing in the Fourier transform region. The echo canceller 13 outputs the transform domain signal group 1 and the transform domain signal group 2 to the spectrum subtraction unit 16.

FIG. 23 is a block diagram showing a configuration example of the echo canceller shown in FIG.

[0192] The echo canceller 13 shown in Fig. 23 has a configuration including a Fourier transformer 35, an adaptive filter group 38, an inverse Fourier transformer 36, a Fourier transformer 37, and a multiplier 39m (m = l to M). is there.

[0193] The far-end signal input from the terminal 31 is expanded in the Fourier transform domain by the Fourier transform 35, and is output to the adaptive filter group 38 for each frequency domain. In addition, the subtraction result input to the subtractor 4 force shown in FIG. 22 via terminal 33 is also expanded in the Fourier transform domain by Fourier transformation, and each multiplier 39m (m = l to M ) Is output.

Multiplier 39m (m = 1 to M) multiplies the signal received from Fourier transform 37 by the voice detection result received via terminal 34, and outputs the calculation result to adaptive filter group 38.

[0195] The adaptive filter group 38 includes M adaptive filters, and receives the signal group 2 output from the Fourier transformer 35 and the signal group 1 output from the multiplier 39m (m = l to M). Then, the adaptive signal is processed using the corresponding signal. The filter output obtained by the adaptive filter processing is output to the inverse Fourier transform 36.

The inverse Fourier transform 36 performs an inverse Fourier transform process on the filter output processed by the adaptive filter group 38 and outputs the processing result from the terminal 32. The signal output from terminal 32 is the output signal for the echo canceller.

In addition, the echo canceller 13 outputs the output signal of the Fourier transform 37 used in the spectral subtraction unit 16 from the outer output terminal 41 as the transform domain signal group 1, and the adaptive filter group 38 The output is output from vector output terminal 42 as transform domain signal group 2. [0198] Transform domain signal group 1 is a signal obtained by Fourier transforming the output signal of subtractor 4 shown in FIG. 22, and transform domain signal group 2 is output from echo canceller 13 shown in FIG. Can be interpreted as a Fourier-transformed signal.

Next, the configuration and operation of the spectrum subtraction unit 16 shown in FIG. 22 will be described with reference to the drawings.

FIG. 24 is a block diagram showing a configuration example of the spectrum subtraction unit shown in FIG.

[0201] The spectral subtraction unit 16 shown in Fig. 24 is the first in that the Fourier transform 60 and the Fourier transform 61 shown in Fig. 11 are deleted and the transform domain signal group 1 and transform domain signal group 2 are input. This is different from the spectral subtraction unit 6 used in the echo suppressor of one embodiment.

[0202] As described above, the transform domain signal group 1 is a signal obtained by subjecting the output signal of the subtractor 4 shown in FIG. 22 to Fourier transform, and the transform domain signal group 2 is the echo canceller 13 shown in FIG. Can be interpreted as a Fourier-transformed signal. These signal groups are exactly the same as the two signals input to the Fourier coefficient subtractor 66m (m = 1 to M) included in the spectral subtraction unit 6 shown in FIG. Therefore, the spectral subtraction unit 16 shown in FIG. 24 outputs the same signal as that of the spectral subtraction unit 6 shown in FIG. Therefore, the echo suppressor of the fifth embodiment shown in FIG. 22 has the same effect as the echo suppressor of the first embodiment of the present invention.

[0203] In the echo suppression apparatus of the fifth embodiment, the spectral subtraction unit 16 is supplied with the transform domain signal group 1 and the transform domain signal group 2 output from the echo canceller 13 to the spectral subtraction unit 16. 16 Fourier transform processing can be reduced.

[0204] Such a configuration can also be applied to the echo suppressors shown in the second to fourth embodiments. In addition to the Fourier transform region, a cosine transform region or the like can be used.

[Sixth embodiment]

In the first to fourth embodiments, an example using the linear echo canceller 3 has been described. For the echo suppression, for example, the subband region echo canceller described in Non-Patent Document 4 is used. It is also possible. In that case, if the processing of the spectral subtraction unit 6 and the spectral suppression unit 7 is performed in the subband region, the filter for conversion to the subband region can be omitted.

FIG. 25 is a block diagram showing the configuration of the sixth embodiment of the echo suppressor of the present invention.

[0206] The echo suppression apparatus of the sixth embodiment performs processing by the echo canceller and the vector subtraction unit in the subband region.

[0207] As shown in FIG. 25, in the echo suppressor of the sixth embodiment, the output signal of the microphone 1 is developed into N frequency bands by the subband analysis filter bank 91, and the far-end signal is subband analyzed. The filter bank 92 expands into N frequency bands.

[0208] The echo canceller unit 93n, the subtractor 94n, the voice detection unit 95η, and the spectral subtraction unit 96η (where η = 1 to Ν) are developed by the subband analysis filter bank 91 and the subband analysis filter bank 92. Corresponding to the specified frequency band.

[0209] The output signal of the spectral subtraction unit 96η is inversely transformed to the original signal region by the subband synthesis filter bank 99 and output as a near-end signal.

[0210] In each frequency band, the processing of the subtractor 94η, the voice detection unit 95η, and the spectral subtraction unit 96η (where η = 1 to Ν) is performed by the number of echo canceller taps and the Fourier transform of the spectral subtraction unit. Except for the difference in scale, the operation is the same as the echo suppressor of the first embodiment shown in FIG. Therefore, description of the configuration and operation of these devices is omitted.

[0211] In the echo suppressor of the sixth embodiment, since all processing is performed in the subband region, the synthesis filter bank in the linear echo canceller 3 and the subband analysis filter bank in the spectral subtraction section are omitted. it can. Therefore, the amount of computation corresponding to the subband analysis filter bank and the subband synthesis filter bank can be reduced, and further, the delay time corresponding to the computation can be shortened.

The configuration of the sixth embodiment shown in FIG. 25 can also be applied to the echo suppression devices shown in the second to fourth embodiments. It is also possible to use a cosine transform region in addition to the Fourier transform region.

[Seventh embodiment] FIG. 26 is a block diagram showing the configuration of the seventh embodiment of the echo suppressor of the present invention.

[0213] The echo suppressor of the seventh embodiment performs echo canceller and spectral subtraction processing in the Fourier transform domain.

As shown in FIG. 26, in the echo suppressor of the seventh embodiment, the output signal force S of the microphone 1 is expanded into M frequency bands by the S Fourier transform ^^ ₁₉₁ , and the far-end signal is transformed by the Fourier transform192. Expanded to M frequency bands.

[0215] Echo canceller 193m, subtractor 194m, voice detector 195m, and Fourier coefficient subtractor 66m (m = l to M) correspond to the frequency band expanded by Fourier transform 191 and Fourier transform 192 Be prepared.

[0216] The output signal of the Fourier coefficient subtractor 66m for each frequency band is inversely transformed to the original signal region by the inverse Fourier transformer 199 and output as a near-end signal.

[0217] The processing of the subtractor 194m and the voice detection unit 195m in each frequency band is the same as the echo suppression apparatus of the first embodiment shown in Fig. 8 except that the number of taps of the echo canceller is different. Works like this. Therefore, descriptions of the configuration and operation of these devices are omitted.

[0218] The echo suppression apparatus of the seventh embodiment performs the processing of the echo canceller and the spectral subtraction unit in the conversion domain in the same manner as the sixth embodiment! In order to perform processing in the region, the number M of frequency bands is larger than in the sixth embodiment, and differs from the echo suppressor of the sixth embodiment in that a Fourier coefficient subtractor 66m is used instead of the spectral sub-translation part. ing.

[0219] In the echo suppression apparatus of the seventh embodiment, since processing is performed in the Fourier transform region, it is not necessary to perform Fourier transform for the processing of the vector subtraction. Therefore, in the seventh embodiment, the Fourier transform and inverse Fourier transform ^^ included in the spectrum subtraction unit are not required, and only the Fourier coefficient subtractor 66m performs the operations necessary for processing the spectrum subtraction! / RU

[0220] In the echo suppression apparatus of the seventh embodiment, the amount of calculation corresponding to the omitted Fourier transformer and inverse Fourier transformer can be reduced.

The configuration of the seventh embodiment shown in FIG. 26 is the echo suppression shown in the second to fourth embodiments. It is also applicable to the device. It is also possible to use a cosine transform region in addition to the Fourier transform region.

[0222] In the seventh embodiment, a linear echo canceller is used. However, a nonlinear echo canceller can also be used for echo suppression. Even in this case, the same effect as described above can be obtained if the processing of the spectral subtraction part and the spectral subtraction part is performed in the Fourier transform domain.

[0223] The echo suppressor of the present invention has been described above using a hands-free telephone as an example. It can be applied to various devices in which loudspeaker loudspeaker and microphone sound pickup are performed simultaneously, such as when echoes from the sky are a problem.

Claims

The scope of the claims

[1] An echo suppression method for suppressing an echo generated by acoustic coupling between a sound collector and a loudspeaker,

When either the output signal of the sound collector or the output signal power of the sound collector is the first signal, and the output signal of the echo canceller is the second signal. ,

An echo suppression method for correcting the first signal by using a leakage coefficient, which is a preset value, used to calculate the leakage amount of the second signal leaking into the first signal.

[2] Dividing the first signal into predetermined frequency regions,

The echo suppression method according to claim 1, wherein the first signal is corrected using the leakage coefficient corresponding to each frequency region.

[3] The eco-suppression method according to claim 1 or 2, wherein a leakage coefficient used for correcting the first signal among a plurality of predetermined leakage coefficients is selected according to a predetermined use situation. .

[4]

Power or amplitude of the output signal of the echo canceller, power or amplitude of the far-end signal

The echo suppression method according to claim 3, wherein the power or amplitude of a specific frequency component of the far-end signal is a deviation.

[5] The usage status is

4. The echo suppression method according to claim 3, which is a volume setting value of the loudspeaker.

[6]

4. The echo suppression method according to claim 3, wherein the positional relationship between the sound collector and the loudspeaker is a relative position.

[7]

4. The echo suppression method according to claim 3, wherein when at least one of the sound pickup unit and the loudspeaker is present, the sound pickup unit or the loudspeaker is used.

[8] Estimating the amount of echo contained in the first signal from the coefficient and the second signal, The echo suppression method according to any one of claims 1 to 7, wherein the first signal is corrected by subtracting the estimated amount of echo from the first signal.

[9] A ratio of near-end speech included in the first signal is estimated from the leakage coefficient, the first signal, and the second signal, and the estimated ratio is multiplied by the first signal. The echo suppression method according to claim 1, wherein the first signal is corrected.

[10] An echo suppressor for suppressing an echo generated by acoustic coupling between a sound collector and a loudspeaker,

An echo canceller that generates an echo replica signal that simulates the echo, and an output signal of the sound collector or an output signal force of the sound collector. When the output signal of the echo canceller is the second signal, a leakage coefficient that is a preset value used for calculating the leakage amount of the second signal that leaks into the first signal is A coefficient generation unit to generate, a correction unit to correct the first signal using the leakage coefficient generated by the coefficient generation unit;

Echo suppression device having

[11] An echo suppressor that suppresses echo generated by acoustic coupling between a sound collector and a loudspeaker.

An echo canceller that generates an echo replica signal that simulates the echo, and an output signal of the sound collector or an output signal force of the sound collector. When the output signal of the echo canceller is a second signal, a frequency dividing unit that divides the first signal for each predetermined frequency region;

For each frequency region of the divided first signal, a leakage coefficient, which is a preset value, is used to calculate the leakage amount of the second signal that leaks into the first signal. A coefficient generator to perform,

A correction unit that corrects the first signal using the leakage coefficient generated by the coefficient generation unit for each frequency region of the first signal; An echo suppressor comprising: a frequency synthesizer that synthesizes the first signal corrected for each frequency domain.

[12] An echo suppressor that suppresses echo generated by acoustic coupling between a sound collector and a loudspeaker,

A transform domain echo canceller that generates an echo replica signal simulating the echo; and an output signal power of the sound collector. A signal obtained by subtracting an output signal of the transform domain echo canceller When the signal divided for each frequency is the first signal and the signal for each frequency domain before inverse linear transformation in the transform domain echo canceller is the second signal, the signal is divided for each frequency domain of the first signal. A coefficient generator for generating a leakage coefficient, which is a preset value, used to calculate the leakage amount of the second signal leaking into the first signal;

A correction unit for correcting the first signal for each frequency domain using the leakage coefficient generated for each frequency domain by the coefficient generator;

An echo suppressor comprising: a frequency synthesizer that synthesizes the first signal corrected for each frequency domain.

[13] An echo suppressor that suppresses echo generated by acoustic coupling between a sound collector and a loudspeaker.

An echo canceller that generates an echo replica signal that simulates the echo, a subband analysis filter that develops an output signal of the sound collector and an output signal of the loudspeaker for each predetermined frequency region,

Either the output signal of the sound collector developed in the frequency domain or the output signal power of the sound collector developed in the frequency domain is a signal obtained by subtracting the output signal of the echo canceller as the first signal. When the output signal of the echo canceller is the second signal, it is used to calculate the amount of leakage of the second signal that leaks into the first signal for each frequency region of the first signal. A coefficient generator for generating a leakage coefficient having a preset value;

A plurality of corrections corresponding to the frequency domain for correcting the first signal for each frequency domain using the leakage coefficient generated for each frequency domain by the coefficient generator. And

An echo suppression apparatus comprising: a subband synthesis filter that synthesizes the first signal corrected by the correction unit.

[14] An echo suppressor that suppresses an echo generated by acoustic coupling between a sound collector and a loudspeaker.

An echo canceller that generates an echo replica signal simulating the echo; a Fourier transform that expands the output signal of the sound collector and the output signal of the loudspeaker for each predetermined frequency region;

A plurality of correction units corresponding to the frequency domain for correcting the first signal for each frequency domain using the leakage coefficient generated for each frequency domain by the coefficient generator;

An inverse Fourier transformer for synthesizing the first signal corrected by the correction unit;

Echo suppression device having

[15] The coefficient generator is

The echo suppression apparatus according to any one of claims 10 to 14, wherein a leakage coefficient used for correcting the first signal is selected according to a predetermined use situation among a plurality of leakage coefficients set in advance.

[16] The coefficient generator is

A detection unit for detecting a predetermined usage state;

A plurality of preset leakage coefficients are stored for each frequency region of the first signal and corresponding to the usage situation, and the leakage coefficient corresponding to the usage situation detected by the detection unit. For each frequency domain; The echo suppressor according to claim 10, comprising:

[17]

17. The echo suppression device according to claim 15, which is any one of power or amplitude of an output signal of the echo canceller, power or amplitude of a far-end signal, or power or amplitude of a specific frequency region of the far-end signal.

[18]

The echo suppressor according to claim 15 or 16, which is a volume setting value of the loudspeaker.

[19]

The echo suppressor according to claim 15 or 16, wherein the relative position relationship between the sound collector and the loudspeaker is set.

[20]

17. The echo suppressor according to claim 15 or 16, which is a sound collector or a loudspeaker to be used when a plurality of at least one of the sound collector or the loudspeaker is present.

[21] The correction unit includes:

A request for correcting the first signal by estimating the amount of echo contained in the first signal from the coefficient and the second signal, and subtracting the estimated amount of echo from the first signal. The echo suppressor according to Item 10 to Item 20!

[22] The correction unit includes:

By estimating the ratio of the near-end signal included in the first signal from the leakage coefficient, the first signal, and the second signal, and multiplying the first signal by the estimated ratio, 21. The echo suppressor according to claim 10, wherein the echo suppressor corrects the first signal.

[23] The correction unit includes:

A first smoothing unit that smoothes an amount according to the amplitude or power of the first signal; a second smoothing unit that smoothes an amount according to the amplitude or power of the second signal; and A first multiplier that multiplies the amount smoothed by the smoothing unit by the leakage coefficient; a subtractor that subtracts the multiplication result of the first multiplier from the amount smoothed by the first smoothing unit; A divider for dividing the subtraction result of the subtractor by the amount smoothed by the first smoothing unit; a third smoothing unit for smoothing the division result of the divider;

23. The echo suppression apparatus according to claim 22, further comprising: a second multiplier that multiplies the first signal by an amount smoothed by the third smoothing unit.